Detection of Visual Impairments in the Pixel Domain

In this article we investigate the utilization of the corrupted IP packets (indicated by checksum) at H.264/AVC decoders. The position of the error within a packet has to be pre-localized by a syntax analysis. The impairments remaining after syntax analysis are further detected in the pixel domain, by means of a voting system, using difference frames and knowledge about the artifact appearance. The blocks considered as erroneous are concealed. The proposed detection of artifacts in the pixel domain considerably improves the probability of error detection achieved by the syntax analysis. The combined system clearly outperforms the typically employed slice discarding in terms of quality measured by mean squared error.